CONVERSATION / EVIDENCE / HUMAN REVIEW

A second opinion.
Not a verdict.

Optional Groq-powered chat and report review. Local AutoMod keeps working separately.

Starting a conversation

Mention Guardify, reply to its message, or start with Guardify, hello. The same member can continue without a mention for up to 90 seconds. Another member’s message or messages addressed to someone else stop implicit follow-ups. Ambiguous standalone greetings do not start a conversation.

This is conservative routing, not perfect intent detection. Guardify does not send autonomous follow-up messages. Replies are plain Discord messages; no hidden command execution or moderation powers are granted to AI.

Temporary memory and personal control

History is per server, channel and member: the latest 12 message entries, at most 1,000 histories in memory. Inactive histories expire after 20 minutes when another eligible chat arrives. No persistent AI chat database is created. /forget clears your temporary history immediately.

/ai_optout enabled:True disables new processing of your messages, clears temporary history and excludes your cached messages from AI report evidence. Existing provider records and already submitted reports are not deleted.

The /report workflow

  1. Run /report user:@member, optionally adding reason. The receipt is private.
  2. Guardify checks that reporting is enabled and a usable log channel exists.
  3. It considers up to 12 eligible target messages from the last 15 minutes, already in its global 200-message cache, from this channel only. You and the bot need access and Read Message History. It does not fetch other channels or download files.
  4. If configured and quota permits, Groq receives the allegation, message excerpts and a small set of filter settings. Instructions inside evidence are treated as untrusted content.
  5. Moderators receive the advisory review and evidence links. A compact report is stored as a local case. Missing evidence or AI failures lead to manual review, not a fabricated finding.
No automatic punishment. AI reports never apply a ban, kick, warning or timeout. The reported member is not declared guilty merely because a report exists.

Configuration and quotas

/ai_settings requires Manage Server. It controls chat, reporting, optional chat channel and daily limits. Defaults: 100 chat requests and 10 AI report reviews per server, reset at UTC midnight. Attempts may consume the local quota even if the provider fails. Provider limits and billing are separate; there are no automatic retries.

The bot owner adds GROQ_API_KEY and a supported GROQ_MODEL privately to the bot’s .env. The site never asks for these secrets. /dashboard includes an AI & Reports overview.